Primitives + workflow modules + templates

Primitives are the bounded, well-defined services PursuitWorks ships and maintains. You don't build these. You call them.

• RFP shred and compliance matrix extraction.
• Document parsing (PDF / DOCX) with sanitization.
• Retrieval-backed memory across project Pursuit Vault and org-wide Capture Library, with provenance citations.
• Audit trail (stage activity, agent runs, retrieval logs, enterprise audit events).
• Governance policy enforcement (rate limits, monthly token budgets, agent allow-lists, role restrictions).
• Durable job queue with retries, idempotency, and stuck-job recovery.
• Storage with signed URLs and Supabase Auth + RLS.

Each module declares its inputs, outputs, allowed agents, and approval lifecycle. Modules expose pages, APIs, and persistence keyed by a stable id. Active modules today.

• Opportunity Match (SAM.gov live discovery, fit scoring against Capture Library context, and promotion into a governed pursuit with the source trail intact).
• Bid / No-Bid decision (8-factor rubric with evidence + recommendation engine).
• Capture Intel Intake (agency, incumbent, customer priorities, teaming posture).
• RFP Shred is the only stage that runs as a multi-step LLM-backed agent loop (tool-use, OODA trace, retry budget).
• Competitive Intelligence, Clause Analyzer, Win Themes, Price-to-Win, Proposal Questions, Teaming Agreement, Past Performance, Proposal Template, Full Proposal. Every one of these stages drafts via Claude through a single streamed completion, gated by approval and deterministic-fallback rules.
• Color-Team Review (Pink/Red/Gold with section-M-factor assignments + finding reconciliation).
• Section L/M Traceability matrix.
• Submission Readiness with FAR flow-down acknowledgments and four-role signoffs.
• Pursuit Debrief feeding org-wide historical_outcomes and proposal_performance_metrics.

Templates compose modules into pursuit shapes. Customers pick a template per pursuit (or define their own) and can disable optional modules pursuit-by-pursuit. Required modules cannot be disabled.

• Services Pursuit (the default 10-module flow).
• IDIQ Task Order (slim, skips most strategy modules since the vehicle is already won).
• SBIR / STTR Phase II (research-heavy).
• Custom templates. Wire your own module order and required gates.

Two AI surfaces, governed separately. (1) Stage drafting. Every stage 1 through 10 produces its initial draft by streaming a Claude completion through the same governed pipeline (prompt-budget compaction, deterministic fallback on restricted content, monthly token budget, distributed rate limits). (2) The agent registry. One real LLM-backed agent (RFP Shred Agent, multi-step tool-use loop) and four deterministic governance checks (Compliance Check, Red Team Audit, Readiness Score, Gate Plan) that audit artifacts using rules, so they never call a model and they can’t hallucinate findings. All five emit OODA-shaped traces, and every entry in the registry is labeled with whether it invokes a model.

• Per-org AI policy controls agents_enabled, allowed_stages, monthly_token_budget, role_restrictions, and background execution.
• Per-plan defaults for requests/min, concurrent runs, and monthly token budgets.
• Stage-level approval lifecycle (draft, committed, approved) with separate review_state (needs_review, changes_requested, approved).
• Submission gate enforces required signoffs before package download.

The proposal-ops mistake we hear most often is bundling six different shapes of work into one AI tool that does one of them well. PursuitWorks pulls those workflows apart, gives you a primitive layer to build on, and a workflow inventory advisor that will sometimes recommend you NOT use PursuitWorks for a particular workflow. The wrong tool for the shape of work is more expensive than no tool at all.